World to Code: Multi-modal Data Generation via Self-Instructed Compositional Captioning and Filtering

TL;DR

This paper introduces World to Code, a pipeline that generates high-quality multi-modal data by converting visual information into Python code, improving VLM performance on various benchmarks.

Contribution

The paper presents a novel multi-modal data generation pipeline that uses VLMs to produce and filter code-based representations, enhancing data quality for VLM training.

Findings

Improves VQA and visual grounding benchmarks

Demonstrates superior cross-modal understanding with code parsing

Shows high-quality data generation via filtering strategies

Abstract

Recent advances in Vision-Language Models (VLMs) and the scarcity of high-quality multi-modal alignment data have inspired numerous researches on synthetic VLM data generation. The conventional norm in VLM data construction uses a mixture of specialists in caption and OCR, or stronger VLM APIs and expensive human annotation. In this paper, we present World to Code (W2C), a meticulously curated multi-modal data construction pipeline that organizes the final generation output into a Python code format. The pipeline leverages the VLM itself to extract cross-modal information via different prompts and filter the generated outputs again via a consistency filtering strategy. Experiments have demonstrated the high quality of W2C by improving various existing visual question answering and visual grounding benchmarks across different VLMs. Further analysis also demonstrates that the new code parsing ability of VLMs presents better cross-modal equivalence than the commonly used detail caption ability. Our code is available at https://github.com/foundation-multimodal-models/World2Code.